Re: OO, C++, and something much better!

[dewar@merv.cs.nyu.edu (Robert Dewar)]

Don Harrison said

"Yes, these mechanisms are inherently unsafe but their purpose is to allow
different views of the same data. There are different ways of acheiving
that and some ways are safer than others. For example, UNCHECKED_CONVERSION
is safer than overlaying because the data is copied thus protecting the
original object."


That is a misconception. Unchecked_Conversion does not require the data
to be copied. The whole point of 13.9(12) is to remove this requirement:

12   An implementation may return the result of an unchecked conversion by
reference, if the Source type is not a by-copy type.  In this case, the
result of the unchecked conversion represents simply a different (read-only)
view of the operand of the conversion.



The design principle here is that since this is a d0-it-at-your-own-risk
and make-sure-you-know-what-you-are-doing operation, it is inappropriate
to waste time trying to increase the safety of the operation. If you
want to ensure that a copy is made, you must make the copy. Note that
in the common case of:

   a := unchecked_convert (b);

the assignment makes a copy anyway (and this is the case where avoiding
the requirement for unchecked_conversion to make a copy of its own 
improves efficiency -- well to be fair the compiler can optimize this
case since it is transparent, but encouraging return by reference will
generally improve efficiency, and is unlikely to be noticeable.

GNAT does take advantage of this permission where appropriate, and UC
for large objects works by pointer punning if the alignments are
compatible (if the alignments don't match, or more specifically if the
target has a stricter alignment than the source, then of course you have
no choice but to make a copy.